适用于即时翻译系统的改进仿射投影算法

Improved affine projection algorithm and its application in instant translation system

针对即时翻译系统应用中存在双端对讲干扰和模型噪声的问题,提出了一种适用于便携式即时翻译系统的改进变步长仿射投影算法。新算法在收敛步长中引入近端信号能量统计量和滤波器收敛程度统计量,根据统计量的改变实时调整步长参数,防止算法发散。仿真结果表明,与传统自适应滤波算法和改进仿射投影算法相比,所提出的算法不但可以有效克服双端对讲干扰,而且在收敛速度、稳态失调等方面也有明显改善。

To solve the problem of double-talk interference and model noises in instant translation system application,this paper proposed an improved variable step-size affine projection algorithm. Introducing the near-end energy estimator and filter convergence statistics in convergence step,it adjusted the step parameters in real-time as the change of the statistics to prevent the algorithm from diverging. Compared with the traditional adaptive filtering algorithms and improved affine projection algorithm,the simulation results show that the proposed algorithm not only efficiently overcomes the interference of double-talk,but also achieves faster convergence speed and lower final misalignment.